AI Glossary: Human-AI Interaction & UX Design

This is probably the most important part of the glossary for designers and these patterns and principles are directly relevant to product design practice. Understanding them enables the creation of AI experiences that are useful, usable, and appropriately trusted.

Interaction Patterns

Conversational UI

Interface paradigm where interaction occurs through natural language dialogue. Reduces learning curves by leveraging innate communication abilities but isn't universally optimal—works best for search, recommendations, and Q&A rather than complex editing.

Reference: Murad, C., Munteanu, C., Clark, L., & Cowan, B.R., "Let's Talk about CUIs: Putting Conversational User Interface Design into Practice", CHI EA 2021

Chat Interface

Visual presentation displaying conversation exchanges in sequential, message-based format with input fields and message bubbles. While familiar, chat-only interfaces create "flow" problems separating AI from existing workflows—embedded approaches often prove more effective.

Reference: Brandtzaeg, P.B. & Følstad, A., "Chatbots: Changing User Needs and Motivations", Interactions, ACM 2018

Prompt-Response Paradigm

Fundamental LLM interaction model where users provide prompts and receive generated responses. Unlike traditional GUIs with predefined options, this shifts creative burden to users who must articulate intent. Creates discoverability challenges requiring thoughtful onboarding and example-driven guidance.

Reference: OpenAI, "Text Generation", OpenAI API Documentation, 2024
Note: Practitioner-defined term emerging from LLM APIs.

Turn-Taking

Conversational protocol governing when each participant contributes, signaled through typing indicators and visual cues. Poor turn-taking design creates awkward interactions where users are unsure if the system is processing or waiting.

Reference: Sacks, H., Schegloff, E.A., & Jefferson, G., "A Simplest Systematics for the Organization of Turn-Taking for Conversation", Language 50(4), 1974

Multi-Turn Conversation

Interactions spanning multiple exchanges where AI maintains awareness of prior messages. Enables follow-up questions without repeating context. Designers must consider visual conversation continuity, context limits, and conversation branching.

Reference: Yi, Z. et al., "A Survey on Recent Advances in LLM-Based Multi-turn Dialogue Systems", 2024

Context Management

Handling accumulated interaction information—conversation history, documents, preferences. LLMs have finite context windows constraining processable information. Consider summarization strategies, selective history inclusion, and explicit context indicators.

Reference: Packer, C. et al., "MemGPT: Towards LLMs as Operating Systems", 2023

Conversation Memory

Ability to retain information across sessions with the same user, enabling personalization and continuity. Designers must balance useful recall against privacy concerns, providing transparency about storage and clear user controls.

Reference: Packer, C. et al., "MemGPT: Towards LLMs as Operating Systems", 2023

System Prompts

Hidden developer instructions shaping AI behavior, personality, and constraints before user interaction. Establish the AI's "character," knowledge boundaries, and guardrails. Understanding system prompts is essential—they fundamentally determine capability and communication.

Reference: OpenAI, "Prompt Engineering", OpenAI API Documentation, 2024

User Prompts

End-user input requesting specific outputs. Range from simple questions to detailed specifications. Designers should minimize "burden of prompting" through templates, auto-completion, and guided inputs helping users express intent.

Reference: OpenAI, "Best Practices for Prompt Engineering with the OpenAI API", 2024

AI Product Design Concepts

AI-First Design

Philosophy where AI capabilities are the primary interaction paradigm rather than an added feature. Rethinks entire user journeys around AI's unique capabilities, requiring consideration of probabilistic nature and error potential from earliest design stages.

Reference: Nielsen, J., "AI: First New UI Paradigm in 60 Years", Nielsen Norman Group, 2023

Human-in-the-Loop (HITL)

Design pattern integrating humans into AI processes to guide, correct, validate, or approve outputs at critical points. Creates feedback loops where corrections improve systems. Implementing HITL means creating clear checkpoints and efficient review interfaces.

Reference: Mosqueira-Rey, E. et al., "Human-in-the-loop machine learning: a state of the art", Artificial Intelligence Review 56, 2023

Automation Bias

Cognitive tendency to over-rely on automated recommendations, accepting AI suggestions without critical evaluation. Paradoxically, the more accurate a system performs, the more likely users trust it blindly when it fails. Designers must counteract through friction points and verification prompts.

Reference: Goddard, K., Roudsari, A., & Wyatt, J.C., "Automation bias: a systematic review", JAMIA 19(1), 2012

Appropriate Trust

Calibrating user confidence to match actual system reliability—neither over-trusting nor under-trusting. Requires transparent limitation communication, consistent performance, and uncertainty feedback. Help users develop accurate mental models through progressive capability disclosure.

Reference: Lee, J.D. & See, K.A., "Trust in Automation: Designing for Appropriate Reliance", Human Factors 46(1), 2004

Mental Models of AI

Internal representations users construct to understand how AI works and what it can do. Users often anthropomorphize AI, assume perfect memory, or expect consistent behavior. Effective design shapes accurate mental models through onboarding and explicit communication.

Reference: Rutjes, H., Willemsen, M., & IJsselsteijn, W., "Considerations on Explainable AI and Users' Mental Models", CHI'19 Workshop, 2019

Expectation Setting

Deliberately communicating capabilities, limitations, and appropriate uses. Microsoft's HAI Guidelines emphasize "making clear what the system can do" at interaction start. Prevents frustration, reduces inappropriate reliance, and builds trust.

Reference: ⚠️ No single authoritative reference — concept appears across UX and HCI literature; related to trust calibration (Lee & See, 2004).

Graceful Degradation

Ensuring AI provides value even below optimal conditions—when confidence is low, context insufficient, or errors occur. Systems should fall back to simpler behaviors, acknowledge limitations, or escalate to human assistance rather than failing silently.

Reference: Saridakis, T., "Design Patterns for Graceful Degradation", TPLoP, Springer 2009

Fallback Behaviors

Predefined actions when normal fulfillment isn't possible—clarifying questions, offering alternatives, graceful declining. Prevent dead-ends and maintain trust. Map all failure modes and create specific responses, avoiding generic "I can't help" responses.

Reference: ⚠️ No single authoritative reference — practitioner-defined pattern in chatbot design; related to graceful degradation.

Error Handling in AI

Strategies for managing, communicating, and recovering from AI errors—incorrect outputs, misunderstood intent, system failures. Unlike traditional software, AI errors are often partial and harder to detect. Effective handling includes easy correction mechanisms and undo capabilities.

Reference: Nielsen, J., "10 Usability Heuristics for User Interface Design" (Heuristic #9), Nielsen Norman Group, 1994

User Experience Factors

Latency Perception

How users psychologically experience response time (differing from objective wait time). Nielsen's research: 100ms feels instant, 1 second maintains flow, 10 seconds risks attention loss. Use streaming and progress indicators to manage perceived latency.

Reference: Nielsen, J., "Response Times: The 3 Important Limits" (from Usability Engineering), Nielsen Norman Group, 1993

Streaming Responses

Displaying output incrementally as generated rather than waiting for completion. Dramatically reduces perceived latency—users see first tokens within milliseconds even if full responses take seconds. Creates engaging "watching the AI think" experience.

Reference: OpenAI, "Streaming", OpenAI API Documentation, 2024

Typing Indicators

Visual cues ("AI is thinking...") communicating active processing. Reassure users the system hasn't crashed and set wait expectations. Sophisticated indicators might show processing stages: "Searching documents...", "Generating response..."

Reference: Gnewuch, U., Morana, S., Adam, M.T.P., & Maedche, A., "'The Chatbot is typing...' – The Role of Typing Indicators in Human-Chatbot Interaction", Pre-ICIS Workshop 2018

Confidence Indicators

Signals communicating system certainty level—percentage scores, verbal qualifiers, or color-coding. Help users calibrate trust and know when to verify. Avoid precision theater—false precision implying more certainty than warranted.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction", ACM CHI 2019

Uncertainty Communication

Patterns transparently conveying when AI is unsure, has limited information, or makes assumptions. Microsoft's HAI Guidelines recommend "scoping services when in doubt." Use hedging language, present alternatives, cite sources, invite verification.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction", ACM CHI 2019

Source Attribution

Citing sources informing AI responses, enabling verification and trust-building. Products like Perplexity popularized in-line citations. Increases credibility, enables fact-checking, helps distinguish synthesis from hallucination.

Reference: Do, H.J., Ostrand, R., Weisz, J. et al., "Facilitating Human-LLM Collaboration through Factuality Scores and Source Attributions", ACM CHI 2024

Explainable Outputs

Responses including reasoning or justification, not just final answers. Helps users verify correctness, learn from AI, and build mental models. Consider layered explanations—brief justifications with expandable detail.

Reference: Zhao, H., Chen, H. et al., "Explainability for Large Language Models: A Survey", ACM TIST 2024

User Control

Principles ensuring users can invoke, dismiss, correct, and override AI behaviors. Microsoft's HAI Guidelines: make AI "easy to invoke when wanted and easy to dismiss when not." Include opt-in/opt-out, customization, and manual overrides.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction", ACM CHI 2019

Reversibility

Ability to undo, modify, or roll back AI actions without consequences. Reduces interaction risk, encourages experimentation. Implement version history, undo mechanisms, confirmation dialogs—treat AI outputs as drafts.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction" (Guideline #12), ACM CHI 2019

Design Patterns

Prompt Templates

Pre-structured formats helping users craft effective prompts through frameworks or fill-in-the-blank structures. Reduce cognitive burden and improve output quality. Offer contextual templates for common use cases.

Reference: MacNeil, S. et al., "Prompt Middleware: Mapping Prompts for Large Language Models to UI Affordances", 2023

Guided Prompting

Walking users through providing information step-by-step rather than requiring complete prompts upfront. Improves accessibility for novices, ensures consistent input quality, captures structured data chat interfaces miss.

Reference: White, J. et al., "A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT", Vanderbilt University, 2023

Few-Shot Examples in UI

Providing 1-3 example input-output pairs to help users understand formats and guide AI behavior. Dramatically improves output quality. Surface this power through "example conversations" or editable templates.

Reference: Brown, T. et al., "Language Models are Few-Shot Learners", NeurIPS 2020

Regeneration

Requesting new outputs for the same input when initial responses are unsatisfactory. Acknowledges AI's probabilistic nature. Include visible controls, preserved alternatives, and potentially adjustable parameters between generations.

Reference: Microsoft HAX Toolkit, "Guidelines for Human-AI Interaction" / HAX Design Library, 2019-2024

Editing AI Outputs

Allowing users to modify or collaborate with generated content rather than accept/reject wholesale. Make outputs clearly editable, track human vs. AI contributions, support hybrid workflows where AI drafts and humans polish.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction" (Guideline #12: Support efficient correction), ACM CHI 2019

Feedback Mechanisms

Systems for rating or reporting output quality, contributing to improvement and monitoring. Include thumbs up/down, detailed forms, error reporting. Make submission frictionless and explain how feedback is used.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction" (Guideline #15: Encourage granular feedback), ACM CHI 2019

Correction Workflows

Structured processes for identifying errors and providing corrections improving future responses. Include highlighting incorrect sections and "teach the AI" modes. Should feel empowering, creating partnership between users and system.

Reference: Microsoft HAX Toolkit, "Design Patterns - When Wrong", 2022

Emerging Interaction Paradigms

Agentic UX

Design approaches for AI operating with greater autonomy—planning actions, using tools, making decisions with minimal per-step input. Requires trust delegation, progress visibility, and intervention points. Balance autonomy against control and transparency.

Reference: Microsoft Design, "UX design for agents: Agent UX Design Principles", 2024

Copilot Patterns

AI as always-available assistant embedded within existing workflows, suggesting improvements without taking over. GitHub Copilot pioneered this; Microsoft extended it across productivity tools. Key considerations: contextual awareness, non-intrusive suggestions, seamless acceptance/rejection.

Reference: Microsoft, "Creating a dynamic UX: guidance for generative AI applications", Microsoft Learn, 2024

AI Assistants

Conversational systems helping accomplish tasks through dialogue, from simple Q&A to complex workflows. Unlike single-purpose tools, assistants are general-purpose and user-directed. Consider personality consistency, capability boundaries, and human support handoff.

Reference: Amershi, S., Weld, D. et al., "Guidelines for Human-AI Interaction", ACM CHI 2019

Autonomous Agents

Systems accomplishing goals independently, making decisions and taking actions without constant direction. Represent a spectrum from "suggest and wait" to "act first, report later." Critical considerations: goal specification, progress monitoring, intervention mechanisms, accountability.

Reference: Wang, L., Ma, C., Feng, X. et al., "A Survey on Large Language Model based Autonomous Agents", 2023

Tool Use in UI

Patterns for AI interacting with external tools, APIs, and services beyond text generation. Enables real actions, not just suggestions. Communicate available tools, show when they're used, display outputs transparently, handle failures gracefully.

Reference: Microsoft, "Tool Use Design Pattern", AI Agents for Beginners Course, 2024

Retrieval-Augmented Interfaces

Experiences built on RAG architecture, where responses are grounded in specific document collections. Show source citations, enable drilling into citations, indicate when AI is synthesizing vs. quoting.

Reference: Lewis, P. et al., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks", NeurIPS 2020

Multimodal Interactions

Interfaces accepting and generating multiple content types—text, images, audio, video. Manage mixed-media flows, indicate available modalities, create coherent experiences across radically different input/output types.

Reference: Wakid, M. et al., "Multimodal Interaction, Interfaces, and Communication: A Survey", MDPI Multimodal Technologies and Interaction, 2024

This glossary is part of a series covering AI and LLM concepts for product designers. Terms without authoritative references are noted for tracking.